Mask R-CNN hangs during training using custom made tfrecords

Accelerated Computing Intelligent Video Analytics TAO Toolkit
20

Your tfrecords file does not contain any ‘image/object/mask’.

More, your create_coco_record.py is different from the create_coco_tf_record.py inside the TLT 2.0 docker. Please refer to it and modify

flags.DEFINE_boolean(‘include_masks’, False,

to

flags.DEFINE_boolean(‘include_masks’, True,

Then, generate tfrecord files via following command. Note, below command just a reference. It only generate training tfrecords files.

PYTHONPATH=“tf-models:tf-models/research” python create_coco_tf_record.py --train_image_dir=./your_images_folder --train_object_annotations_file=./IRUV_train_v1.json --output_dir=./result --train_caption_annotations_file=dummy_file.json